The Impact of Editorial Slant: Evidence from
the Hearst Media Empire
Siona Listokin∗
School of Public Policy
George Mason University
Jason Snyder
School of Law and the Kellogg School of Management
Northwestern University
∗ For useful comments and suggestions we thank the anonymous referees, Rui DeFigueiredo,
Yair Listokin, David Nasaw, and seminar participants at UC Berkeley. Jason Snyder would
like to thank the Searle Foundation for support. All mistakes are ours alone. Corresponding
author is Siona Listokin ([email protected])
1
Abstract
We examine whether editorial slant influences electoral outcomes in the
context of one of the most powerful media conglomerates in US history.
In the early 1900s, the Hearst newspaper empire was politically charged
and considered influential. We test if the Hearst newspapers affected
elections. Using a difference-in-differences and matching methodology, we
find that the introduction of a Hearst newspaper into a county did not
change electoral outcomes compared to similar counties - in contrast to
other studies of media effects. We consider explanations for the results,
and offer historical perspective to an issue that remains both salient and
ambiguous.
2
Does the media affect our political opinions and behavior? Social scientists have
struggled with this question since the rise of a “mass media” in the 19th century
that utilized new technologies to reach large audiences.1 Conventional wisdom
places newspapers, radio and television as critical determinants of certain historical events; thus, it has long been debated whether or not the Spanish-American
War in 1898 was ignited by the journalism of William Randolph Hearst and
Joseph Pulitzer, and that John F. Kennedy beat Richard Nixon in the 1960
presidential election in part because of his better appearance on television.2
More recently, the rise of powerful media conglomerates has renewed the concern about media effects and the possibility that the political preferences of the
owners of news companies influence popular perceptions of policies and politics.
Researchers have attempted to determine what influence, if any, the news
media has on public opinion. In a series of studies in the 1940s and 50s, political scientists Lazarsfeld, Berelson and Gaudet (and later, Katz) investigated
the influence of the mass media on presidential elections by examining survey
answers and correlations to news availability, media habits and personal preferences (Lazarfeld et al 1940; 1944; 1955). An entire field of political science
research has since been devoted to the subject, and recently economists have
contributed to the study of media effects.3 This literature is a reflection of both
the importance of the research questions on media effects and the difficulty of
properly identifying them since media exposure can be both a cause and a consequence of political preferences. While the results are mixed, the empirical
evidence does favor the possibility that editorial slant and media effects can
1 For example, the Frankfurt School in the 1930s critiqued the emergence of a "mass media
culture"; more formally, Lazarsfeld, Berelson and Gaudet (1940, 1944) studied survey data to
determine the relationship between the media and public opinion.
2 For example, Campbell (2003) opens his book on "yellow journalism" detailing the prevailing view of media’s influence on the Spanish-American War. Druckman (2003) reviews
the Kennedy-Nixon debates.
3 A literature review follows. See (McCubbins 1986) for a review of earlier literature; (Kull
et al 2003; Zaller 1996; Gentzkow and Shapiro 2006; Druckman and Parkin 2005; George and
Waldfogel 2005; Della Vigna and Kaplan 2007).
3
wield critical influence on election outcomes.
We examine media effects in the context of the William Randolph Hearst
newspaper empire in the early 20th century. Hearst owned one of the most
powerful media conglomerates in United States history, and his newspapers
had a distinct editorial slant. Did the Hearst newspapers affect presidential
and congressional election outcomes between 1880 and 1938? We choose this
historical context because the Hearst newspaper empire is unique in its scope
and in the nature of its expansion. In the early 20th century, Hearst built and
expanded a network of newspapers and magazines in cities across the United
States; by 1930, his newspapers were read by almost 20% of the country’s total
population (Nasaw 2001). Hearst was also a political figure, having served two
terms as a Democratic congressman, and remained powerful in the Democratic
Party during the era. Furthermore, Hearst’s editorial bias was reflected in both
news articles of his newspapers and in the prominent placement of opinion pieces
on the front page of the newspapers, following the practice of the era. These
features allow us to examine the effects of editorial slant on electoral outcomes
with limited concern of measurement error that plagues studies that focus on
more recent elections, where media outlets are diffuse and political bias in news
reporting is relatively low.
The results of our study suggest that there is no Hearst effect on voting
outcomes and voter turnout; this result is robust to a number of specifications.
We use a difference-in-difference methodological approach that compares voter
results and turnout in metropolitan areas where a Hearst newspaper was introduced to those that never received a Hearst paper. We infer from our results
that the presence of a Hearst newspaper in an MSA was not associated with a
change in election outcomes.4 Considering the unexpected nature of our results
4 That
is, the parameter estimate is tightly centered around zero, despite being statistically
insignificant.
4
– especially given the recent empirical research suggesting strong media effects
on election outcomes – we consider a number of possibilities that may have produced this result aside from an absence of media effects. Nevertheless, this is
an empirical illustration of the limits of editorial slant influence on real voting
outcomes.
Editorial Slant and Elections
Our focus is on the extent to which a powerful news media outlet influences
Presidential and Congressional election outcomes. Political news media can
influence voters by acting as the intermediary between voters and their elected
representatives (Paletz, 1999). "Media effects" has thus been a concern for social
scientists because of the informational power it confers on a private industry. In
their seminal analyses of the media and public opinion, Lazarsfeld, Berelson and
Gaudet reported that media messages have a fairly small, but significant, effect
on political opinions (Lazarsfeld et al 1945). These results were supported by
much of the succeeding literature, with many studies conducted using National
Election Survey data (Zaller 1996; Kull et al 2003).5
Inevitably, treatment of media effects leads to consideration of bias in the
news media, as those editorial slants distort the media’s informational function.
Much of the research consideration has focused on first measuring media bias
and then using survey data to determine its influence on voters (Dalton et al
1998; Gilens and Hertzman 2000; Kahn and Kenny 2002; Druckman and Parkin
2005; Gentzkow and Shapiro 2006). Some recent studies have shifted focus away
5 The difficulty of isolating media effects on voter attitudes and behavior (and minor results)
cast a certain degree of doubt on these results, leading one researcher to declare in 1993 that
"The state of research on media effects is one of the most notable embarrassments of modern
social science" (Bartels, 1993).
5
from content analysis and polling data, exploiting plausibly exogenous variation
in media expansion to test how the media affects voting outcomes against the
counterfactual in the absence of the media.
Our study resembles a paper by DellaVigna and Kaplan that looks at the
effect of the introduction of the Fox News channel on the 2000 presidential
election (DellaVigna and Kaplan 2007). The authors are able to exploit the
rapid expansion of the cable news channel in towns throughout the country, and
they find that the introduction of the Fox News channel convinced 3 to 8 percent
of voters to vote Republican. Gentzkow similarly studies the rapid expansion of
television in the 1950s and tests its influence on voter turnout (Gentzkow 2006).
The paper finds that turnout dropped, possibly because television featured less
election coverage than other news sources that viewers substituted away from.
George and Waldfogel test if the expansion of The New York Times in the 1990s
changed local voting and newspaper readership patterns (George and Waldfogel
2006). They find that the expansion of the Times decreased local newspaper
circulation and among certain populations, decreased voting in local elections.6
In studying the effects of the Hearst newspaper empire on electoral outcomes, we employ similar methodological tools as the papers above. Additionally, we consider the explicit media bias of the Hearst newspapers (though we
do not quantitatively measure it). The Hearst newspaper empire remains unparalleled in its combination of market power and political bias.7 Informally,
if the modern-day media effects measured in previous papers are applicable to
Hearst’s era – and we are able to isolate them – there would be a potentially
substantive example of changes in election outcomes in the first quarter of the
6 (Oberholzer-Gee
and Waldfogel 2001; Stomberg 2004)
unscientific consensus elevates Hearst’s media empire as paradigmatic of the media’s
ability to influence elections and policy. For (an unscientific) example, see the Wikipedia entry
on “Media Effects” or the Encyclopedia Britannica’s entry for William Randolph Hearst. The
most widely adopted elementary school (upper division) social studies textbooks (according to
the American Textbook Council) at least partly ascribe the Spanish-American War to Hearst’s
reporting (Harcourt/Holt “Social Studies”)
7 An
6
1900s. Beyond historical curiosity, the Hearst papers’ strong presence represents
a unique opportunity to measure media influence on elections.
The Hearst Newspapers
William Randolph Hearst took control of his first newspaper in 1887, in his
home town of San Francisco. Hearst started his practice of reporting national
news in city papers and contracted with the New York Herald to cable articles
to his San Francisco Examiner. This move “changed the face of the Examiner
and journalism on the West Coast,” and the competition folded or followed suit
in introducing national news coverage (Nasaw 2001).
During the next 25 years, Hearst started newspapers in more than seventeen
cities, starting with the largest metropolitan areas (New York and Chicago), and
then delving into smaller cities with geographic and demographic diversity. As
the Hearst newspapers expanded its market scope, the number of readers and
percent of market share also grew. Hearst papers held at least 10% market share
in their markets, with an average of about 25% share of major newspapers after
five years. At their height, 20 million Americans read a Hearst newspaper daily.
For comparison, The New York Times had a total circulation of 1.1 million in
2006, and an average of 1.5 million Americans watched the Fox News Channel
during primetime in August 2006 (0.9 million watched CNN).
The locations and timing of the Hearst expansion were not random. Although Hearst was ostensibly a multimillionaire, his money was tightly controlled by his mother, Phoebe Hearst, and shifts in their relationship – most
radically, Phoebe’s death – often precluded newspaper “buying sprees.” Thus,
the periods between 1911-12 and 1919-22 saw the acquisition of four and ten
more newspapers, respectively, as Phoebe granted more assets to Hearst in 1911
and following her death in 1919.
7
Hearst chose the cities in which to expand for different reasons. Table 1 lists
the cities and years where Hearst introduced a newspaper.8 The San Francisco
Examiner was a gift from his father. The choice to move into New York was
strategic: Hearst meant to build a publishing empire while establishing himself
in the Democratic Party leadership, and the New York newspaper market was
necessary for both objectives (Nasaw 2001). In 1900, the Democratic National
Committee explicitly asked Hearst to start a paper in Chicago in order to increase their exposure through sympathetic reporting in the mid-West. After his
early expansion into the large population centers (New York, Chicago, Boston,
Detroit, Baltimore), Hearst later expanded into some smaller cities, such as
Atlanta and Syracuse, to extend his geographic scope into new regions. His
deliberate growth in upstate New York in the early 1920s was also attributed
to his desire to run for the governorship of New York.9
As mentioned, there are two aspects unique to the Hearst newspaper empire
and this time period that make the Hearst papers an interesting and highly
relevant empirical study of the effects of media bias: the overt subjectivity
injected into the Hearst papers and their overwhelming market penetration.
It may be hard for today’s newspaper readers to appreciate the level of bias
injected into newspaper reporting during this time. Gentzkow et al describe
newspapers during the 19th century as “often public relations tools funded by
politicians,” though they emphasize that bias declined during the early 20th
century (Gentzkow et al 2004).
Hearst generally bought small or ailing newspapers, and upon purchase
Hearst and his editors changed the news content, most notably by tapping
into the Hearst editorials, cartoons and articles that were wired through private cables and syndicated across the country (Nasaw 2001). The content of
8 In our analysis, we consider Hearst’s initial entry into a metropolitan newspaper market.
In some instances, there were multiple Hearst publications.
9 http://timesunion.com/specialreports/tu150/
8
these Hearst features were generally urban, worker oriented and geared to readers with Democratic Party leanings. This did not mean, however, that Hearst
monolithically towed the Democratic Party line during the almost forty years
under study in this paper; to the contrary, Hearst had frequent “lapses” away
from the Democratic mainstream in favor of western progressive politics and the
lower taxes offered by Republicans in the post-World War I era (Nasaw 2001).
Nonetheless, Hearst was comfortable using his newspapers as publicity tools
for his own campaigns, to support other candidates, to ingratiate himself to the
Democratic Party and even to promote his other commercial endeavors. According to a Hearst biographer, "Hearst employed the power of the media to set
the national political agenda, [as] a muckraking progressive trustbuster (Nasaw,
xiv)," and ran his papers as "pro-labor, pro-immigrant and anti-Republican.
(106)" Hearst’s first newspaper, the San Francisco Examiner was "defiantly prolabor, anti-capital and anti-railroad," positions well-aligned with the Democratic
Party. Hearst himself viewed editorial campaigning as a necessary function of
his newspapers. Hearst wrote in 1936, "The word crusade must not be considered as an attack... It is essential for the papers to conduct constructive
campaigns for the benefit of the community with which they are associated.
And it is of vital importance for the papers to identify themselves with the
aims and ambitions of the community and to make themselves recognized forces
in the accomplishment of these aims and ambitions (The Newspaper Credo of
William Randolph Hearst)."10
This involvement in political campaigns is most evident in presidential and
congressional election periods. In the 1896 presidential election, "The Journal
10 Hearst
was less pugnacious in official communications with his editors and reporters. In
1933, Hearst ordered that the following bulletin be posted in his newsrooms: "Be fair and
impartial. Don’t make a paper for Democrats and Republicans, or Independent Leaguers.
Make a paper for all the people and give unbiased news of all creeds and parties. Try to do
this in such a conspicuous manner that it will be noticed and commented upon." (from The
Newspaper Credo of William Randolph Hearst)
9
office became headquarters for the [William Jennings Bryan] campaign... His reporters followed Bryan’s every step along the campaign trail; interviews and articles were published daily (Nasaw 2001)."11 In the summer of 1931, Hearst gave
explicit orders to his editors to use news reporting and cartoons to paint President Herbert Hoover as treasonous and a presidential failure (Carlisle 1965).
Hearst’s newspapers were important elements in local congressional elections,
too. Between 1928 to 1932, a number of congressman received editorial support
and favorable coverage in Hearst newspapers in their respective congressional
districts (or themselves wrote articles and editorials for Hearst newspapers), including William Borah of Idaho, Hiram Johnson of California, Thomas Walsh of
Montana and George Norris of Nebraska.12 While this support was notable even
for its time, it is unclear whether the presence of a Hearst paper systematically
affected election results.
Measuring the Hearst Effect
We measure the effect of the Hearst newspapers by examining the conglomerate’s expansion over time and across different cities. We examine the electoral
outcomes before and after a Hearst newspaper enters a new county and compare this to electoral outcomes in similar counties that did not receive a Hearst
newspaper during the relevant era. We first use the conventional difference in
differences OLS estimator:
(1) Yi,t = βHearsti,t +
!
i
Countyi +
!
t
Y eart + εi,t
This specification examines how the outcome Yi,t (for instance the percent11 Hearst’s control over tone and bias became notorious just before World War II broke out,
when his papers were highly sympathetic to Hitler and Missolini. Both men wrote "reports"
from their home countries for Hearst publications.
12 For example, the San Francisco Examiner expressed support for Hiram Johnson and his
policies on June 19, 1928, pg 26; June 21, 1928, pg 30. Johnson wrote a public thank you in
the paper on August 29, 1928, pg 2.
10
age of the electorate in county i, in year t, that voted for the Democratic Congressional candidate) changed when Hearst entered the county (Hearsti,t is a
dummy variable equal to one if a Hearst newspaper has entered the market)
conditioning on county fixed effects, Countyi , and year fixed effects, Y eart .
This approach eliminates any fixed differences across counties and over time.
Throughout all of our specifications the error terms are clustered at the level of
the metropolitan statistical area13 to account for auto-correlation in the data
across counties and over time. This clustering relaxes the assumption of independence of the error terms of counties that are in close proximity to one
another. Panel applications that do not account for autocorrelation in the data
can dramatically underestimate the standard error for the parameters of interest
(Bertrand, et al 2004).
Though difference in differences is a widely used methodology for the analysis
of panel data, it often relies on serious assumptions about the trends prior
to an intervention for causal inference to be valid. For example, if prior to
the introduction of a Hearst newspaper the trends in voting outcomes differ
significantly between a treatment county (the county where Hearst enters) and
control counties (counties where Hearst didn’t enter), then using a conventional
difference in differences estimator could lead to a spurious inference that Hearst
ownership had an influence of voting outcomes. For example in Table 2A we
observe that the trends in the percentage of votes for Democrats over the four
elections prior to the introduction of a Hearst paper differ from the potential
control counties. There is a pronounced decline in the average votes received by
Democratic congressional candidates in the areas that Hearst expanded into (for
example, in column 1 of Table 2A, Democratic congressional vote share declines
from 0.50 to 0.45). To partially address these concerns we include a yearly
linear trend at the county level to account for differences in the pre-intervention
13 A
metropolitan statistical area covers multiple counties.
11
trends. This specification is given by:
!
!
!
(2) Yi,t = βHearsti,t + i Countyi + t Y eart + i Countyi ∗ T imet + εi,t
Here the time trend T imet is a counter that starts at zero in the year 1880.14
The main effect of the linear time trend is absorbed by the year dummies, so
the interaction term controls for county specific time trends that are not picked
up by the main year effects. This technique is standard in the difference in
differences literature and has been employed in a variety of settings (Wolfers
2006).
We further address these concerns by create control groups that match the
treatment group to a set of controls that have similar pre-trends and are of
similar size.15 We construct a series of sets W (1)...W (K), each of which consist
of a single treated county and 4 control counties for all treated counties K.16
The objective in constructing W (k) is that they find the "best" set of control
counties for the treatment county k. To construct a set of controls we proceed
in a two step process. First we require that a potential control match the
treatment exactly on the voting population bin.17 We then employ a standard
distance metric18 that allows for us to find a set of potential control counties
based on observable pre-intervention voting outcomes. For a potential control
observation x, a treatment observation z, and a diagonal matrix V consisting
of the inverse of the variances of the observations we use the following distance
metric:
14
#1/2
"
!
(3) "z − x" = (z − x) V (z − x)
T imet = 0 in year 1880, T imet = 1 in year 1881, etc.
15 (Card
1990) uses a similar approach
this paper the estimation procedures will be done with replacement. That is
being a control for one treated county does not preclude said county from being a control for
another treated county.
17 We construct three bins: counties with less that 25,000 voters, counties with 25,000 to
100,000 voters, and counties with over 100,000 voters.
18 This is standard in many matching applications. We could also use the Mahalanobis
metric as a means of computing the distance (Abadie et al 2004)
16 Throughout
12
From here we can construct the set W (i) by taking the 4 closest control
counties. All of methods were implemented in Stata (Abadie, et al 2004).
In order to evaluate the final outcomes, the conventional differences in differences estimator can be used on each group W (k). For each sub-group W (k),
specification (1) is run on the treatment and its best controls, no longer using
the less desirable controls for the estimation. To recover the average treatment
effect on the treated across all counties we run the following model using dummies Groupk for each k group from W (k) (i.e. each for each Hearst county and
its matches):
(4) Yi,t = βHearsti,t +
!
i
Countyi +
!
t
Y eart +
!
k
Groupk ∗ Y eart + εi,t
Here Hearsti,t is the average treatment effect on the treated. Notice how the
variable Hearsti,t is the only variable that is not interacted. One could potentially interact this with Groupk to recover the estimates for each treatment and
its m controls. This would be the exact same result as running each regression
separately within each group. The usefulness of using specification (4) is that it
allows for the easy inclusion of matching methods within a standard regression
framework.
Historical Data & Sample Selection
Hearst Data. We obtained data on Hearst expansion cities from historical and biographical studies. Occasionally, years of the newspaper expansions
differed slightly (one year off), in which case we relied on the David Nasaw biography. Any county that is in a Metropolitan Statistical Area where a Hearst
paper is introduced is considered to be a treated observation.
Voting Data. The voting data comes from the ICPSR "Electoral Data for
Counties in the US: Presidential and Congressional Races, 1840-1972." This
is our source for the dependent variables, Democratic vote shares from the
13
congressional elections over this time period, Democratic vote share from the
presidential elections over this time period, and turnout for congressional elections during this time period. We use the ICPSR Democratic Party vote share
"equivalent," which accounts for party variations in the 18th and 19th century.
This data also includes population numbers that are computed to represent the
number of eligible voters.
Unfortunately since this is historical data, there are often numerous missing
observations. We find that on average missing observations come from smaller
counties19 and are less frequent over time.20 We also find that the number of
missing observations is approximately the same between counties where a Hearst
paper was present and ones where Hearst did not enter.
Our sample covers the years 1880 − 1938. We decided on this sample since
it is a natural historical era, the time between the end of the reconstruction and
the start of World War II. We only study counties that are in large Metropolitan
areas.21
Empirical Results
The difference in differences OLS estimates from equation (1) & (2) are
shown in Table 1. The dependent variables are Democratic Party vote share
and voter turnout for both presidential and congressional elections. We regress
each dependent variable on the explanatory variable of the Hearst entry in the
MSA, a population control and year and county fixed effects. Each regression
is also run with an additional control for the linear year trend and county interaction from equation (2), which are in the even numbered columns. The
19 In 1900 the average county population where the observations were missing was 7,434
while the average county size without missing observations was 16,394.
20 Approximately 22% of the obeservations were missing in 1900, while in 1932 only 12%
were missing.
21 Summary statistics can be found in an online appendix, available upon request.
14
results show that Hearst entry has a very small and statistically insignificant
effect on voting outcomes and turnout. Despite the statistical insignificance,
the results are informative. The confidence intervals for each parameter estimate are tightly centered around zero (+/- less than 0.10), indicating a precisely
estimated zero.22
Turning to the matching results, we first examine the effectiveness of the
matching process in Tables 2A-2D. While pre-trends and levels of the unmatched
sample are in general close to the treated counties, after matching we are able
to obtain matches that are extremely close to the pre-trends of the treated
counties. Additionally the matching process enables us to obtain much closer
matches in terms of county population.
Tables 3-6 show the results using the matched samples, with the congressional and presidential vote share, and congressional and presidential voter
turnout in the four tables. In the matching estimates we restrict the estimates
to looking for the Hearst effect 10 years after Hearst entered into a city. Furthermore we eliminate counties where over 20% of the sample is missing. The
matching results show a very small and statistically insignificant "Hearst effect"
on all of the four vote outcome variables. As with the difference in difference
regression, most of the matching estimates are closely centered around zero.
We examine the robustness of the matching results in column (2) of Tables
3-6 by only looking at counties that are near the center of the MSA. We find
the results are unchanged. In column (3) we make the sample more restrictive
by only admitting counties with less than 10% of the data missing. Again the
results are unchanged. Finally in column (4) we linearly impute the missing
data in these counties and find the results unchanged.
22 In
the online appendix we show that the results are robust to including interpolated data
and cubic linear time trends.
15
Discussion
The results of our empirical test are strongly suggestive that the Hearst
newspaper empire did not change election vote shares or turnout in the early
20th century, and provide a thorough empirical illustration of the limitations
of media bias. The coefficient estimates for the “Hearst” dummy variable are
consistently insignificant, through a host of specifications intended to address
potential problems in the identification. This stability mitigates our concerns
that our failure to reject the null hypothesis is a false finding. We also note that
the data set used in this paper have produced reasonable, significant results
in past research (for example, Ansolabehere et al 2001; Carson and Roberts
2005). The coefficients in our results are small in scale, especially considering
the baseline of the MSA’s in the years before Hearst’s entry. We conclude that
while we cannot reject the null hypothesis of no effect, we can infer that the
Hearst newspapers had a “zero” effect on election outcomes on congressional
races, presidential races, and turnout.
The consistency of these results, which suggest no Hearst influence on voting
outcomes, prompt the subsequent question: Given Hearst’s market penetration
and editorial message, why wasn’t there a "Hearst effect?" We consider three
types of plausible explanations that are consistent with the historical record and
our results: (1) The treatment effect of Hearst’s entry into a media market was
in fact very small, (2) There was a substantial influence upon the electorate,
however the effect washed out, or (3) that Hearst did not influence electoral
voting, but he did influence government along other dimensions such as policy
making or the selection of politicians.
The first possibility, that Hearst did not influence election results because
his newspapers did not change voter opinions, is most obviously consistent with
our baseline results. This possibility was addressed indirectly in a 1936 poll in
16
Fortune Magazine. The poll asked "Do you think the influence of the Hearst
papers upon national politics is good or bad?" The results are shown in Table
7.
[Insert Table 7 about here]
The consistency of the favorable response suggests that a large portion of
Hearst newspaper readers were members of a stable population that generally
agreed with Hearst, in which case the presence of a Hearst publication would
not have an influence on their opinions and voting outcomes.
It is also important to note that at the height of Hearst’s circulation (and
market share), Hearst was somewhat more inconsistent in his support for the
Democratic Party. The Hearst newspapers were favorable to Hoover when he
ran against Democrat Al Smith. In the congressional elections of 1930, the
papers did not follow a national policy, and the California papers abstained from
political support of particular congressmen. Since our empirical test includes a
number of periods in the mid and late thirties, the lack of effect may reflect this
inconsistency in political campaigning (Carlisle, 1965).
The second possibility, that the Hearst effect exists but is not observable, is
conceivable for the same reasons we might have expected the Hearst papers to
have a discernible influence on elections. The Hearst papers may have changed
underlying features of treatment cities that negate Hearst’s influence. It is possible that Hearst changed the distribution of voters’ preferences without changing
the actual electoral outcomes. For example, suppose that following the introduction of a Hearst newspaper in a metropolitan area, local newspaper competitors
responded with similarly powerful pro-Republican articles. If voters were uniformly distributed along a unidimensional continuum of policy preferences and
after the entry of Hearst the distribution of voter preferences was transformed
into a bimodal distribution with modes at both the left and the right end of the
17
spectrum, the position of the median voter might still remain unchanged. Many
simple models of political polarization would be consistent with this finding of
no change in average voting outcomes.
In order to determine the competitive effect of a Hearst newspaper entry,
we examined the circulation records for US cities collected in 1910, 1919 and
1929. These records provide market information in Hearst cities over time; given
the varied dates of Hearst entry, we can observe competitive influences both
immediately after a Hearst purchase and decades later. The circulation numbers suggest that the smaller newspaper markets like Pittsburgh and Milwaukee
turned into “two paper” markets, which would support the possibility that other
newspapers competitors strategically placed themselves in political opposition.
In most of the mid-sized and large markets, however, the competitive landscape
was not altered so radically; Hearst newspapers gained market share by cutting
into smaller circulation papers. While the circulation figures cannot reveal the
content of the competing newspapers, it does not appear that local newspaper
competitors strategically responded to Hearst with right-leaning content that
would be reflected in polarized market share figures.
The final possibility that we consider is that the Hearst newspaper empire
did not influence electoral voting, but influenced other electoral and policy outcomes. This would be consistent with our results if, for example, the Hearst
papers influenced policies whose absence would have been discernable in the electoral outcomes. There is no question that political campaigns and presidential
administrations considered the reaction of the Hearst newspapers when choosing
policies. For example, both the Hoover and FDR administration crafted parts
of their labor policies with Hearst in mind (Carlisle, 1965). Communications between Hearst, FDR, Joseph Kennedy (SEC Chairman) and Ed Coblentz (manager of multiple Hearst papers) reveal that Hearst relayed detailed opinions on
18
domestic policy and suggestions for cabinet secretaries. Thus, the Hearst newspaper may have influenced politics by virtue of their communication power, in
manners that would not be reflected in presidential and congressional elections.
Hearst’s papers may also have influenced candidate choice. Hearst was pivotal in getting FDR the Democratic Party’s nomination at the 1932 Democratic
National Convention. The delegates were deadlocked between FDR (Governor,
NY) and Alfred E. Smith (1928 Democratic Presidential candidate) for several
days, while a third candidate, John Garner (Speaker of the House), controlled
the deciding votes. Hearst supported Garner. Eventually, Joseph Kennedy
called Hearst in the middle of the night and convinced Hearst to have Garner drop out of the race and throw his delegates to FDR. Hearst agreed and
FDR won in the next ballot (Nasaw, 2001). Hearst may have greatly influenced
voters’ opinions without a discernible effect on election results, by virtue of a
change in candidates or policies. We cannot reject this possibility, and certainly
these historical facts are supportive of a non-electoral Hearst effect.
In future research it would be of considerable interest to learn whether Hearst
and other media effects influence behavior along other political dimensions. For
instance it might be possible that the newspaper ended up being more issue
oriented, forcing politicians that had an electorate in it’s domain to support
certain issues that Hearst favored. One could imagine studying how Hearst’s
entry (or any other media "message") may have influenced roll call votes, public
spending, or the spread of corruption.23
The Hearst newspapers’ overwhelming circulation percentages alarmed politicians and journalists in the early 20th century; nonetheless, at the time their
influence was unclear. One paper actually noted, "More interesting [than circulation] would be some test of the actual influence of Hearst’s ideas upon his more
than five million daily readers, but no means of investigation seems to uncover
23 This
would be in line with Stromberg (2004)
19
this sort of information (Doan, 1932)." This paper uncovers Hearst’s influence
on elections, and opens the door to the possibility that his papers had other
means of influence. If nothing else, the Hearst newspapers at their height were
a means of political communication that has not been duplicated in the United
States since; the lack of evidence for a strong media effect in this case suggests
that "media effects" may have their most substantial influence on the competitive and political environment leading up to elections. As the study of media
bias and electoral influence continues, our study is indicative of the necessity –
and limitations – of quantitative research to address long-held assumptions.
20
References
[1] Abadie, A, Drukker, D, Herr, J and G. Imbens. Implementing Matching
Estimators for Average Treatment Effects in Stata. The Stata Journal, Vol.
4, No. 3. 2004.
[2] Ansolabehere, S, J. Snyder and C. Stewart. Candidate Positioning in U.S.
House Elections. American Journal of Political Science, Vol. 45, No. 1, 2001.
[3] Bartels, L. Messages Received: The Political Impact of Media Exposure.
The American Political Science Review, Vol. 87, No. 2, 1993.
[4] Bertrand, M., E. Duflo & S. Mullainathan. How Much Should We Trust
Differences-in-Differences Estimates?. The Quarterly Journal of Economics,
MIT Press, vol. 119(1), 2004.
[5] Bovitz, G, James N. Druckman, and Arthur Lupia. When Can a News
Organization Lead Public Opinion? Ideology Versus Market Forces in Decisions to Make News. Public Choice Volume 113(1-2), 2002.
[6] Campbell, W. Joseph. Yellow Journalism: Puncturing the Myths, Defining
the Legacies. New York, Praeger, 2003.
[7] Card, D. The Impact of the Mariel Boatlift on the Miami Labor Market.
Industrial and Labor Relations Review, 1990
[8] Carlisle, R. The Political Ideas and Influence of William Randolph Hearst,
1928-1936. Ph.D. diss., University of California, Berkeley, 1965.
[9] Carson, J and J. Roberts. Strategic Politicians and U.S. House Elections,
1874-1914. The Journal of Politics, Vol. 67, No. 2, 2005.
21
[10] Dalton, Russell and Beck, Paul. Partisan Cues and the Media: Information
Flows in the 1992 Presidential Election. The American Political Science
Review, Vol. 92(1), 1998.
[11] Della Vigna, Stefano and Kaplan, Ethan. The Fox News Effect: Media Bias
and Voting, Quarterly Journal of Economics, Vol. 122, 2007.
[12] Doan, E. Chain Newspapers in the United States. Journalism Quarterly,
December 1932.
[13] Druckman, James N. The Power of Television Images: The First KennedyNixon Debate Revisited. The Journal of Politics, Volume 65, Number 2,
May 2003 , pp. 559-571(13)
[14] Druckman, James and Parkin, Michael. The Impact of Media Bias: How
Editorial Slant Affects Voters. The Journal of Politics, Volume 67(4), 2005.
[15] Gentzkow, M. Television and Voter Turnout. Quarterly Journal of Economics. CXXI (3). August 2006 .
[16] Gentzkow, M, EL Glaeser and C Goldin. The Rise of the Fourth Estate:
How Newspapers Became Informative and why it Mattered. NBER, 2004.
[17] Gentzkow, M and Shapiro, J. Media, Education and Anti-Americanism.
The Journal of Economic Perspectives, Vol 18(3), 2004.
[18] _____. Media Bias and Reputation. Journal of Political Economy. 114
(2). April 2006.
[19] George L. and Waldfogel, J. The New York Times and the Market for Local
Newspapers. American Economic Review, March 2006.
22
[20] Gilens, Martin and Hertzman, Craig. Corporate Ownership and News Bias:
Newspaper Coverage of the 1996 Telecommunications Act. The Journal of
Politics, Volume 62(2), 2000;
[21] Hearst, William Randolph. A Handbook of Journalism. Imprint: Hearst
Corporation, 1963.
[22] Kahn, Kim Fridkin and Kenney, Patrick.The Slant of the News: How Editorial Endorsements Influence Campaign Coverage and Citizens’ Views of
Candidates. American Political Science Review, Vol 96: 381-394, 2002.
[23] Katz, E., and P. F. Lazarsfeld. Personal Influence. New York: Free Press,
1955.
[24] Klapper, J. The Effects of Mass Communication. New York: The Free
Press, 1960.
[25] Kull, Steven, Clay Ramsey, Stefan Subias, Evan Lewis, Phillip Warf. Misperceptions, The Media and the Iraq War. The PIPA/Knowledge Networks
Poll, 2003
[26] Lazarsfeld, P. F., Berelson, B., & Gaudet, H. (1940). Erie County study.
Inter-University Consortium for Political and Social Research. Ann Arbor:
University of Michigan Press.
[27] Lazarsfeld, P. F., Berelson, B., & Gaudet, H. (1944). The people’s choice:
How the voter makes up his mind in a presidential campaign. New York:
Duell, Sloan, & Pearce.
[28] Mayhew, D. The Electoral Connection. Yale University Press, New Haven
CT. 1974.
[29] McCombs, M. The Agenda-Setting Function of Mass Media. The Public
Opinion Quarterly, Vol. 36, No. 2, 1972.
23
[30] Nasaw, D. The Chief: The Life of William Randolph Hearst. Houghton
Mifflin 2001.
[31] Oberholzer-Gee, Felix and Waldfogel, Joel. Electoral Acceleration: The
Effect of Minority Population on Minority Voter Turnout. NBER Working
Paper 8252, 2001.
[32] Paletz, David L. The Media in American Politics: Contents and Consequences. New York, Longman, 1999.
[33] Stromberg, David. "Radio’s Impact on Public Spending." The Quarterly
Journal of Economics. Feb 2004. Vol. 119, Iss. 1; p. 189
[34] Waldfogel, Joel and Lisa George. "The New York Times and the Market
for Local Newspapers." forthcoming, American Economic Review.
[35] Wolfers, Justin. “Did Unilateral Divorce Raise Divorce Rates? A Reconciliation and New Results” American Economic Review. December 2006.
96[5], 1802-1820.
[36] Zaller, John., "The Myth of Massive Media Impact Revived: New Support
for a Discredited Idea," in Political Persuasion and Attitude Change, 2001
24
Table 1: Effect of Hearst Entry on Voting Outcomes
Hearst Entry
County * Time Trend
Population Quadratic
Year Fixed Effects
County Fixed Effects
Observations
(1)
(2)
Congressional Congressional
Democratic
Democratic
Vote Share
Vote Share
-.026
-.009
(.023)
(.016)
N
Y
Y
Y
Y
Y
Y
Y
19466
19466
(3)
Presidential
Democratic
Vote Share
-.019
(.025)
N
Y
Y
Y
11826
(4)
Presidential
Democratic
Vote Share
.006
(.029)
Y
Y
Y
Y
11826
(5)
(6)
Congressional Congressional
Election
Election
Turnout
Turnout
.015
-.002
(.031)
(.013)
N
Y
Y
Y
Y
Y
Y
Y
22397
22397
(7)
Presidential
Election
Turnout
.011
(.033)
N
Y
Y
Y
11681
(8)
Presidential
Election
Turnout
-.010
(.014)
Y
Y
Y
Y
11681
Note: These estimates are obtained through ordinary least squares regressions. This sample follows years 1880 through 1938. Robust standard
errors in brackets clustered at the MSA level.
25
Table 2 A & B: Candidate Vote Share Match Quality
Table 2A: Match Quality for Congressional Democratic Candidate Vote Share
Conngressional Democrat
Candidate Vote Share (t-8)
Conngressional Democrat
Candidate Vote Share (t-6)
Conngressional Democrat
Candidate Vote Share (t-4)
Conngressional Democrat
Candidate Vote Share (t-2)
County Population in
Year
Adoption
(1)
(2)
(3)
Treatment
Unmatched
Control Sample
Matched
Control Sample
.501
.501
.487
(.017)
(.003)
(.008)
.440
.513
.441
(.012)
(.003)
(.006)
.457
.493
.458
(.016)
(.003)
(.008)
.450
.505
.453
(.018)
(.004)
(.009)
56895
22848
47504
(9206)
(779)
(3745)
Table 2B: Match Quality for Presidential Democrat Candidate Vote Share
Presidential Democrat
Candidate Vote Share (t-12)
Presidential Democrat
Candidate Vote Share (t-8)
Presidential Democrat
Candidate Vote Share (t-4)
County Population in
Year
Adoption
Treatment
Unmatched
Control Sample
Matched
Control Sample
.499
.500
.493
(.014)
(.003)
(.006)
.495
.490
.492
(.014)
(.003)
(.007)
.413
.462
.419
(.014)
(.003)
(.007)
56895
22848
47334
(9206)
(779)
(3748)
Note: Standard errors in brackets clustered at the MSA level.
26
Table 2 C & D: Turnout Variables Match Quality
Table 2C: Match Quality for Congression Voter Turnout
Congressional Voter Turnout (t-8)
Congressional Voter Turnout (t-6)
Congressional Voter Turnout (t-4)
Congressional Voter Turnout (t-2)
County Population in
Year
Adoption
Treatment
Unmatched
Control Sample
Matched
Control Sample
.561
.623
.572
(.015)
(.004)
(.008)
.456
.529
.463
(.017)
(.004)
(.008)
.532
.595
.537
(.016)
(.004)
(.008)
.491
.505
.473
(.016)
(.004)
(.008)
56895
22848
49571
(9206)
(779)
(3799)
Treatment
Unmatched
Control Sample
Matched
Control Sample
.595
.651
.605
(.016)
(.004)
(.008)
.601
.652
.615
(.015)
(.004)
(.008)
.579
.625
.582
(.015)
(.004)
(.008)
56895
22848
47297
(9206)
(779)
(3521)
Table 2D: Match Quality for Presidential Voter Turnout
Presidential Voter Turnout (t-12)
Presidential Voter Turnout (t-8)
Presidential Voter Turnout (t-4)
County Population in
Year
Adoption
Note: Standard errors in brackets clustered at the MSA level.
27
Table 3: Influence of Hearst Entry on Congressional Democratic Vote Share
Hearst Entry
Exclude Exterior Areas
More Restrictive Sample
Use Imputed Data
Year Fixed Effects
County Fixed Effects
Observations
(1)
Congressional
Democratic Vote
Share
.007
(.016)
N
N
N
Y
Y
6346
(2)
Congressional
Democratic Vote
Share
.013
(.015)
Y
N
N
Y
Y
4385
(3)
Congressional
Democratic Vote
Share
.007
(.019)
N
Y
N
Y
Y
5383
(4)
Congressional
Democratic Vote
Share
.012
(.017)
N
N
Y
Y
Y
7150
Table 4: Influence of Hearst Entry on Presidential Democratic Vote Share
Hearst Entry
Exclude Exterior Areas
More Restrictive Sample
Use Imputed Data
Year Fixed Effects
County Fixed Effects
Observations
(1)
Presidential
Democratic Vote
Share
.004
(.020)
N
N
N
Y
Y
4143
(2)
Presidential
Democratic Vote
Share
.004
(.021)
Y
N
N
Y
Y
2751
(3)
Presidential
Democratic Vote
Share
.007
(.020)
N
Y
N
Y
Y
3410
(4)
Presidential
Democratic Vote
Share
.002
(.019)
N
N
Y
Y
Y
4150
Note: These estimates are obtained through ordinary least squares regressions on the matched sample. This sample comes from four
election cycles prior to the entry and six elections following entry. Robust standard errors in brackets clustered at the MSA level.
28
Table 5: Influence of Hearst Entry on Congressional Voter Turnout
Hearst Entry
Exclude Exterior Areas
More Restrictive Sample
Use Imputed Data
Year Fixed Effects
County Fixed Effects
Observations
(1)
Congressional Voter
Turnout
.008
(.010)
N
N
N
Y
Y
6975
(2)
Congressional Voter
Turnout
.006
(.012)
Y
N
N
Y
Y
4682
(3)
Congressional Voter
Turnout
.007
(.009)
N
Y
N
Y
Y
6544
(4)
Congressional Voter
Turnout
.009
(.010)
N
N
Y
Y
Y
7150
Table 6: Influence of Hearst Entry on Presidential Voter Turnout
Hearst Entry
Exclude Exterior Areas
More Restrictive Sample
Use Imputed Data
Year Fixed Effects
County Fixed Effects
Observations
(1)
Presidential Voter
Turnout
-.008
(.018)
N
N
N
Y
Y
4099
(2)
Presidential Voter
Turnout
-.002
(.021)
Y
N
N
Y
Y
2718
(3)
Presidential Voter
Turnout
-.009
(.024)
N
Y
N
Y
Y
3379
(4)
Presidential Voter
Turnout
-.007
(.018)
N
N
Y
Y
Y
4120
Note: These estimates are obtained through ordinary least squares regressions on the matched sample. This sample comes from four
election cycles prior to the entry and six elections following entry. Robust standard errors in brackets clustered at the MSA level.
29
Table 7: Fortune Magazine Poll, 1936
Question: Do you think te influence of the Hearst newspapers upon national politics is good or bad?
Areas without Hearst Newspaper
Good
10.7%
Bad
27.6%
Don't Know
61.7%
Areas with at least one Hearst Newspaper
10.5%
43.3%
46.2%
30